Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Registration is a Powerful Rotation-Invariance Learner for 3D Anomaly Detection
Authors: Yuyang Yu, Zhengwei Chen, Xuemiao Xu, Lei Zhang, Haoxin Yang, Yongwei Nie, Shengfeng He
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on the Anomaly-Shape Net and Real3D-AD datasets demonstrate that our method consistently outperforms existing approaches in effectiveness and generalizability. 4 Experiments 4.1 Experimental Settings Datasets. Evaluation is conducted on Anomaly-Shape Net [2] and Real3D-AD [1]. Evaluation metrics. We evaluate anomaly detection performance at both the object and point levels using the Area Under the Receiver Operating Characteristic Curve (AUROC). 4.4 Ablation Study We conduct ablation studies on the registration strategy, training objective, memory bank feature composition, and performance under noisy point clouds to evaluate their respective effects. |
| Researcher Affiliation | Academia | 1South China University of Technology 2Guangdong University of Petrochemical Technology 3Singapore Management University EMAIL EMAIL EMAIL EMAIL EMAIL EMAIL |
| Pseudocode | No | The paper describes the methodology and training objective using descriptive text and mathematical formulations. It also includes a high-level overview diagram (Figure 2) but does not present any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | https://github.com/CHen-ZH-W/Reg2Inv 5. Open access to data and code Question: Does the paper provide open access to the data and code, with sufficient instructions to faithfully reproduce the main experimental results, as described in supplemental material? Answer: [Yes] Justification: The code is available at https://github.com/CHen-ZH-W/Reg2Inv. |
| Open Datasets | Yes | Datasets. Evaluation is conducted on Anomaly-Shape Net [2] and Real3D-AD [1]. Anomaly Shape Net is a synthetic 3D anomaly detection dataset... Real3D-AD is a real-world high-resolution dataset... |
| Dataset Splits | Yes | Anomaly Shape Net is a synthetic 3D anomaly detection dataset with 1,600 samples across 40 categories, each containing 4 normal samples in the training set. Real3D-AD is a real-world high-resolution dataset with 12 object categories, each category has 4 normal training samples and 100 test instances. |
| Hardware Specification | Yes | Experiments are conducted on a single RTX 4090D GPU. Using an RTX 3090 GPU with a batch size of 1, our method achieves an average processing time of 2.53 seconds per Real3D-AD sample, which is faster than other memory bank-based approaches, as shown in Tables 7. |
| Software Dependencies | No | The model is trained for 100k iterations using the Adam optimizer with an initial learning rate of 1e-4. The learning rate is first linearly warmed up for 10k steps, then decayed following a cosine schedule to 10% of the initial value. While the paper mentions the "Adam optimizer", it does not provide a specific version number for this optimizer or any other software libraries or dependencies used in the implementation. |
| Experiment Setup | Yes | The model is trained for 100k iterations using the Adam optimizer with an initial learning rate of 1e-4. The learning rate is first linearly warmed up for 10k steps, then decayed following a cosine schedule to 10% of the initial value. Under our experimental setup (RTX 3090 GPU, batch size 1), full training takes approximately 27 hours on Anomaly-Shape Net (40 categories) and about 34 hours on Real3D-AD (12 categories). |